J. Srinonchat, S. Danaher, J.H. Allen (UK), and A. Murry (New Zealand)
Speaker Dependent Coding System, KSOFM, CELP
There are different optimisation criteria in the design of speech compression algorithms and in development speech coding system. This paper presents a new design of Code Excited Linear Predictive (CELP) base on the federal standard FS1016 CELP coder in a development of a low bit rate speech processing system. This new CELP coder is designed in an environment of Speaker Dependent Coding System (SDCS). In this new CELP coder system, an unsupervised learning neural algorithm is employed to combine with Line Spectral Pairs (LSP) parameters for design a new codebook for reducing the bit rate in transmission. The performance of this unsupervised learning neural algorithm for SDCS will be presented. This technique can be improve the quality of 4.8 kb/s CELP coder in SDCS, and reduce the bit rate around 22% while maintain the quality of synthesis sound.
Important Links:
Go Back
